October 28

Your Daily Note,

AI worlds you can steer, a warning about giving agents access to your money, and fresh investment in companies building at scale.

In today’s Noted:

Google’s Genie 4 creates interactive worlds you can change while they run

Image source: Google

We noted:

Google DeepMind has introduced Genie 4, a world model that generates interactive, increasingly realistic environments with sound, dialogue and real-time changes. Unlike a conventional video generator that produces a predetermined clip, Genie 4 lets users steer a simulated scene using natural language. It is currently a research preview for trusted testers.

Why it matters

Potential applications include interactive entertainment, storyboarding and robotics simulation, although these are not all established commercial uses yet. AI-generated media is moving beyond pictures and clips toward environments people can interact with. That could change how creative teams prototype ideas and how developers test situations before building them in the real world.

Bank of America warns customers not to hand their passwords to AI agents

We noted:

Bank of America has reminded customers not to share their banking usernames and passwords with AI tools or agents that offer to act on their behalf. In an email titled “Protecting your finances in the AI World,” the bank warned that sharing credentials with an outside service could let an agent perform financial actions without the customer’s knowledge.

Why it matters

Expect more questions about how AI tools are authorised to access sensitive accounts. A tool that can navigate a website is not automatically a tool that should be trusted with your money. Use official integrations and permission controls where available, rather than giving an agent your main login. AI agents can take action on a user’s behalf, and when they do, a mistake or security breach can have real consequences.

Disruptive raises a $10 billion fund for big bets on AI and defence

We noted:

Investment firm Disruptive has announced a $10 billion fund intended to back at least 10 late-stage companies, with individual investments of up to $1 billion. The Dallas-based firm, whose previous investments include Palantir, Airbnb and Spotify, is targeting capital-intensive areas including AI and defence.

Why it matters

The fund does not mean every AI startup will suddenly find it easier to raise money, and it is not aimed at early-stage founders looking for their first cheque. Building AI at scale can require substantial spending on computing infrastructure, talent and deployment, and large funds are positioning themselves to finance companies that need to grow far beyond the typical startup budget.

The data centre boom could help modernise the US power grid

We noted:

The growing electricity demand from AI data centres is putting pressure on America’s ageing grid, but industry leaders say the same demand could accelerate investment in new grid technologies. At TechCrunch Disrupt, energy-sector leaders discussed ways to make better use of existing infrastructure and expand capacity.

Why it matters

These technologies are not an instant fix for electricity shortages or rising bills. Their impact will depend on deployment, costs, regulation and how quickly utilities can modernise their systems. AI infrastructure is creating demand beyond chips and servers, and it is also increasing the need for power generation, transmission, energy management and more efficient electrical equipment.

A former Meta AI director wants to replace Office and Google Workspace

We noted:

Rohit Patel, who spent eight years developing AI at Meta, has launched Ndoc, a startup aiming to rethink the workplace suite rather than simply add an AI assistant to existing office software.

Why it matters

Ndoc is still early, so it is not yet a proven replacement for the tools businesses already use. Patel says the product is built using a mix of open-source models. The next stage of workplace AI may not be about adding a chatbot to Word, spreadsheets or slides. It may be about redesigning the workflow so analysis, documents and interactive outputs are created together.

Products worth trying

 

Everything else in AI

AI labs are making competing claims about a major computer science assumption

Following OpenAI’s recent claim that its models found a way to refute the Exponential Time Hypothesis, Anthropic has released its own result. The hypothesis has long helped computer scientists reason about how difficult certain problems are to solve as they grow in size.

The claims are significant, but they still need scrutiny from researchers. Computer scientists are organising a collaborative review process to understand the proofs and what they would mean if validated. This is not the same as proving that all hard computational problems are easy, or that existing encryption has been broken.

 

AllenAI asks whether AI can reliably judge original research

AllenAI has released a benchmark and study examining whether language models can tell if a research idea is genuinely novel. Its tests found that seemingly small changes to a prompt could sharply change a model’s verdict, including a drop in accuracy from 93.5% to 40.9% in one experiment.

The takeaway is important for businesses as well as scientists: an AI system can generate plausible ideas without being reliable at determining whether they are original. More reasoning or research retrieval did not consistently solve the problem.

 

Japan’s GMO Internet investigates a suspected data leak

GMO Internet says an attacker may have published customer and business information after unauthorised access to its Onamae Desktop Cloud service. The company said the material was still being verified, and the full scope of the incident was unclear. It has advised customers to reset passwords and received reports of unauthorised operations involving customer trading accounts.

 

Solid-state transformers could help modernise electricity infrastructure

MIT Technology Review has highlighted solid-state transformers, an emerging class of power equipment that uses power electronics to control electricity conversion and flow. As electricity demand grows and more renewable generation and data centres connect to the grid, technologies that help manage power more flexibly are attracting attention. Their broader impact will depend on cost, reliability and deployment at scale.

 

Retell AI shows how voice agents can handle customer calls

OpenAI’s Retell AI case study describes a no-code platform for building voice agents that can answer questions, schedule appointments, qualify leads and handle administrative tasks. It is an example of AI being used to complete defined customer-service workflows rather than only generating replies. The case study was published in 2025, so it is a useful example of an existing application, not a new launch from today.

See you tomorrow
same time, different note.

~ Stackari